Usage Guide COCO AI May 2026 9 min read

COCO Agent Teams: Zylos Runtime and Architecture | COCO AI

Complete usage guide for the COCO Agent Teams architecture and Zylos open-source runtime. Learn how five-layer memory architecture, omnichannel deployment to Telegram, Lark, and Slack, scheduled tasks, persistent memory, and autonomous orchestration enable specialized AI employees to collaborate on complex enterprise workflows.
CO
COCO AI TeamAI Digital Employee Platform · Product and Content
Published May 28, 2026
Last Updated May 28, 2026
The TL;DR

The Problem

Solo AI models face fundamental limits: context windows constrain complex tasks, they cannot collaborate, and they require manual orchestration. On most platforms, AI employees run in isolation, unable to form synergistic teams for enterprise-grade workflows.

The Solution

COCO Agent Teams architecture delivers specialized role distribution, shared context, real-time collaboration, and autonomous orchestration. Built on the Zylos five-layer memory architecture, AI employee teams accumulate knowledge long-term and collaborate on cross-functional tasks.

The Result

COCO own 6-person team runs 30+ AI agents daily, handling code review, deployment, competitive analysis, and customer support, achieving 10x release efficiency. An MCN agency completes network-wide new media data collection and analysis in under 1 hour.

Explore Other COCO Guides

Before vs After COCO

Before

  • Solo AI: one model handles everything, limited by context window
  • AI tools run in isolation: information silos, no synergy
  • Every new task requires re-providing full background context
  • Manual orchestration of input/output across multiple AI tools

After

  • Agent Teams: specialized roles with clear division of labor
  • Shared context: real-time information flow within the team
  • Persistent memory: AI employees continuously accumulate business knowledge
  • Autonomous orchestration: AI teams auto-coordinate task distribution

Zylos Runtime: COCO Technical Foundation

Zylos is COCO self-developed open-source autonomous AI agent runtime, responsible for managing, scheduling, and orchestrating the complete agent lifecycle. Its core innovation is a five-layer memory architecture designed to outperform alternatives like OpenClaw for long-term enterprise scenarios. Zylos is open source on GitHub with 2,200+ stars and is the flagship among COCO 10+ core open-source projects. It supports persistent memory so AI employees retain interaction history and business knowledge, plus multi-channel communication that unifies messaging across Telegram, Lark, Slack, and more.

Using Omnichannel Deployment: One Dashboard, Full Coverage

COCO Agent Cloud lets you deploy AI employees to Telegram, Lark, Slack, and other channels simultaneously from a single dashboard. All interaction history and context are unified in the backend regardless of which channel the user communicates through. HxA Connect serves as the cross-platform human-AI collaboration real-time message bridge, ensuring seamless message flow across channels.

Using Scheduled Tasks and Persistent Memory for Autonomous AI Operations

All COCO plans starting from Air include scheduled tasks and persistent memory. Scheduled tasks let AI employees auto-execute on defined cycles. One internet company team set competitor tracking, user feedback compilation, and channel data aggregation all to run automatically, with reports ready every morning. Persistent memory means AI employees remember conversation history, business context, and user preferences.

From Single AI Employee to Agent Teams

COCO design philosophy: the future of AI is coordinated professional teams, not solo models. Air and Pro plans start with 1 AI employee. Ultra supports 3 Pro AI employees forming a team. Enterprise supports unlimited instances. The upgrade path is clear: validate value with 1 AI employee, then expand into a specialized Agent Team. COCO own proof: 6-person team runs 30+ AI agents daily at 10x release efficiency.

HxA Suite: Enterprise Toolset

HxA Suite is COCO enterprise-grade efficiency toolkit, including HxA Connect for cross-platform message bridging and HxA Dashboard for Agent Team monitoring. The HxA suite maximizes Agent Team effectiveness with team-level visibility, performance monitoring, and collaboration orchestration. Enterprise plan users can deeply customize HxA Suite capabilities.

Reproducible Setup: Full Walkthrough

Step 1: Clone and Inspect Zylos (Optional, for Developers)

Terminal commands:
git clone https://github.com/coco-xyz/zylos.git
cd zylos && cat ARCHITECTURE.md

This gives you direct access to the runtime source code, memory layer implementation, and benchmark suite. For non-developers, this step is optional — the SaaS Agent Cloud handles all of this automatically.

Step 2: Create Agent Team via Dashboard

Exact navigation: Agent Cloud Dashboard (icoco.ai after login) → Left sidebar: "AI Employees" → Click "+ Create Team" (button at top-right of the employee list) → Name your team → "Add Members" dropdown: select 2-3 AI employees (must already exist as individual employees) → Under "Orchestration Mode", select "Autonomous" (Zylos auto-coordinates) or "Manual" (you define the workflow via the visual editor) → Click "Create Team".

What you see: Team card appears in the dashboard with member list, orchestration mode badge, and a "Test Team" button that sends a sample multi-agent task.

Step 3: Verify Persistent Memory Works

Test procedure: (1) In Telegram/Lark/Slack, tell your AI employee: "Remember that our Q2 target is 15% growth in the APAC region." (2) Wait 5 minutes. (3) In a new conversation, ask: "What is our Q2 target?"

Expected result: AI employee responds with "Your Q2 target is 15% growth in the APAC region."

If it fails: Check the Agent Cloud Dashboard → AI Employee → "Memory" tab → verify the fact appears in the semantic memory layer. If not, check that persistent memory is enabled in the employee settings (toggle under "Capabilities").
Agent Teams 5-Layer Memory Open Source Zylos

Experience the Power of Agent Teams

Deploy specialized AI employees that collaborate, remember, and orchestrate autonomously.

Try COCO Free

Case Study Source Attribution

Customer Testimonials

All customer case studies and quotations referenced in this guide are sourced from the official COCO website (icoco.ai). Testimonials were collected and published with customer consent. Specific metrics cited: State-owned bank: loan due diligence reduction from 3 days to 2 hours (as stated by the branch vice president on icoco.ai). MCN agency: network-wide data collection and analysis in under 1 hour (as stated by the marketing director). Internet company: nearly doubled release cadence after automating competitor tracking and data aggregation (as stated by the team lead). Individual investor: daily automated stock screening and analysis (self-reported). K12 institution: 24/7 AI teaching assistant deployment (as stated by the academic director). Recruitment: doubled hiring efficiency (as stated by the recruitment lead).

Methodology & Transparency

Data Sources

All pricing information, feature descriptions, and customer case studies referenced in this guide are sourced directly from the official COCO website (icoco.ai) and COCO documentation (docs.icoco.ai). Case studies reflect real customer testimonials published on icoco.ai as of May 2026.

Limitations

This guide reflects COCO platform capabilities as of May 2026. Features, pricing, and channel support are subject to change. Always refer to icoco.ai for the most current information. Performance metrics cited in customer testimonials represent specific use cases and may not reflect results in all scenarios.

How We Tested

This guide was compiled by the COCO AI product and content team based on direct platform usage, customer interviews, and official documentation. Pricing and feature tables were verified against icoco.ai as of the publish date. Where customer metrics are cited (e.g., time savings, efficiency gains), the specific customer and context are identified.